diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index 5c9badb0..57c7dfb2 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -68,7 +68,7 @@ jobs:
       timeout-minutes: 60
       run: |
         ./tools/run_in_docker.sh -ni ccache -M 25Gi
-        ./tools/run_in_docker.sh -ni cmake -G Ninja -S . -B build -DCMAKE_CUDA_ARCHITECTURES=80
+        ./tools/run_in_docker.sh -ni cmake -G Ninja -S . -B build
         ./tools/run_in_docker.sh -ni cmake --build build --config ${{env.BUILD_TYPE}} --target all
 
     # - name: benchmark
@@ -79,4 +79,4 @@ jobs:
       timeout-minutes: 10
       working-directory: ${{github.workspace}}/build
       # Execute tests defined by the CMake configuration.
-      run: ../tools/run_in_docker.sh -ni -o --gpus=all ctest --output-on-failure -C ${{env.BUILD_TYPE}}
+      run: ../tools/run_in_docker.sh -ni ctest --output-on-failure -C ${{env.BUILD_TYPE}}
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 761f77df..191df2bb 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -69,7 +69,7 @@ endif()
 
 # set architecture for CUDA
 if (NOT DEFINED CMAKE_CUDA_ARCHITECTURES)
-  set(CMAKE_CUDA_ARCHITECTURES 80)
+  set(CMAKE_CUDA_ARCHITECTURES native)
 endif()
 
 # Build TORCH_CUDA_ARCH_LIST
diff --git a/tools/run_in_docker.sh b/tools/run_in_docker.sh
index 31453dda..53c482af 100755
--- a/tools/run_in_docker.sh
+++ b/tools/run_in_docker.sh
@@ -30,7 +30,7 @@ function usage() {
   echo "Examples:"
   echo "  ${progname} cmake -G Ninja -S . -B build"
   echo "  ${progname} cmake --build build --target all"
-  echo "  ${progname} -ni -d vectorchai/scalellm_devel:cuda12.8 -o '--gpus=all' ctest"
+  echo "  ${progname} -ni -d vectorchai/scalellm_devel:cuda12.8 ctest"
   echo ""
   exit 1
 }
@@ -39,7 +39,7 @@ function usage() {
 
 # Default image
 IMAGE="vectorchai/scalellm_devel:cuda12.8"
-RUN_OPTS=()
+RUN_OPTS=("--gpus=all")
 
 INTERACTIVE=1
 while [[ $# > 1 ]]; do